ByteLite
The file comes back exactly. Or it failed.
I am building ByteLite to represent data in a smaller, self-contained form without guessing, dropping bytes, or hiding reconstruction information somewhere that was not counted.
Internal tests prove exact reconstruction on the current proof artifacts. They do not yet prove the complete self-contained target across broad file corpora.
No hand-waving. No uncounted reconstruction state. No partial credit.
Why this exists.
Storing and moving data costs real money, and most of what gets stored is described far less efficiently than it could be. That part is not controversial. What is hard is doing better without ever guessing — because the moment a system guesses, it is no longer returning your file, it is returning something that resembles it.
ByteLite is an attempt at the honest version of that. ByteLite’s core encoding and decoding mechanism is deterministic. It does not rely on probabilistic inference or confidence scores to decide what the original data probably was.
ByteLite is not built around probabilistic entropy coding. Its research target is a deterministic, recursively self-contained representation architecture.
Three rules I will not trade away.
-
Exact reconstruction
The bytes that come out match the bytes that went in, or the attempt failed. There is no "close enough" and no partial credit.
-
Complete counting
Any size claim is measured from the complete artifact. Moving part of the cost somewhere uncounted is not compression, it is bookkeeping.
-
Deterministic operation
Same input, same result, every time. No confidence scores, no sampling, nothing that needs a retry to land on the right answer.
What is actually proven.
Claims move up only when the evidence behind them moves up. Here is where each one sits today.
Proven internally
- Exact lossless round-trip on internal test artifacts, verified by comparing the rebuilt file against the original.
- Strict artifact accounting: the reported size is counted from the complete artifact, never estimated.
- Deterministic behavior: the same input and build produce the same output.
Not proven
- The universal target. A smaller self-contained representation for every source file is what ByteLite is being built to achieve, not something it has demonstrated.
- A self-contained final artifact. Development currently relies on explicit reconstruction evidence that the final architecture must absorb.
- A specific compression ratio. No public ratio is claimed.
- A completed full enwik9 benchmark result.
- Independent third-party verification. Every result above is internal.
- Production readiness. ByteLite is research under active development, not a shipping product.
What it would change, if it works.
Smaller representations mean less to store and less to move. That is the entire commercial idea, and it is why the licensing model only charges against savings somebody actually measured.
I am not going to put a percentage here. I do not have a verified one, and a number on a marketing page that nobody measured is exactly the kind of thing this site exists to avoid. When there is a real figure, it will arrive with the measurement attached.
Founder access.
Nothing here is finished, so nothing here charges you. Each card says what state it is really in.
- ByteLite Personal - monthly Being configured
- ByteLite Personal - annual Being configured
- Cordel Connect - monthly Being configured
- Cordel Connect - annual Being configured
- Cordel Play - founder reservation Being configured
I am Tash Broadwater. I am building ByteLite in Helena, Montana. There is no hidden engineering department and no marketing team allowed to turn an unfinished result into a finished claim. I am responsible for what this site says, and I can be asked about it directly.
If a result fails, it stays failed until the code proves otherwise.
Ask me something hard.
Technical validation, licensing, partnership and investor conversations all come to me directly. If you can falsify something on this site, that is the most useful message I can get.